Glib2 build problem

[N_Ox]

Le 22 sept. 07 à 00:58, David DeHaven a écrit :

>
> Encountered a similar error trying to install glib2 on a system  
> running Leopard (9A527):
> ld: bc out of range (571728 max is +/-64K) from  
> _g_atomic_int_exchange_and_add in .libs/gatomic.o to  
> _g_atomic_int_exchange_and_add$stub in .libs/ 
> libglib-2.0.0.1400.0.dylib in _g_atomic_int_exchange_and_add  
> from .libs/gatomic.o
>
> There's a bc instruction (or variant thereof, probably the "bne-")  
> that's trying to jump further than 64K bytes from the PC, that's  
> causing the linker error. This would have affected PowerPC builds  
> only.
>
> I poked around a short while and couldn't figure out what's causing  
> the error. It could be a bug in the Leopard preview build of ld64...
>
> I did notice the other functions in gatomic.c didn't have that  
> problem, but they all have a sync instruction before the label, so  
> I dropped a nop in front of the label in  
> g_atomic_int_exchange_and_add (and then later g_atomic_int_add):
>  __asm__ __volatile__ ("nop\n"
> 			".Lieaa%=:       lwarx   %0,0,%3\n"
>
> Which compiles to:
> 	.align 2
> 	.p2align 4,,15
> 	.globl _g_atomic_int_exchange_and_add
> _g_atomic_int_exchange_and_add:
> 	nop
> .Lieaa17:       lwarx   r2,0,r3
>         add     r0,r2,r4
>         stwcx.  r0,0,r3
>         bne-    .Lieaa17
> 	mr r3,r2
> 	blr
>
> And things are building fine now, though wholly untested.
>
> Note the only difference in the assembly output is the nop between  
> the global and the local label (and the tab before the .Lieaa17  
> label, which doesn't make any difference):
> _g_atomic_int_exchange_and_add:
> 	.Lieaa17:       lwarx   r2,0,r3
>
> Otherwise they're the same. This really looks to me like a bug in  
> ld64.
>
> Here's a patch:
> --- old/glib-2.14.0/glib/gatomic.c	2007-09-21 15:32:52.000000000 -0700
> +++ new/glib-2.14.0/glib/gatomic.c	2007-09-21 15:32:33.000000000 -0700
> @@ -276,7 +276,8 @@ g_atomic_int_exchange_and_add (volatile
>  			       gint           val)
>  {
>    gint result, temp;
> -  __asm__ __volatile__ (".Lieaa%=:       lwarx   %0,0,%3\n"
> +  __asm__ __volatile__ ("nop\n"
> +			".Lieaa%=:       lwarx   %0,0,%3\n"
>  			"         add     %1,%0,%4\n"
>  			"         stwcx.  %1,0,%3\n"
>  			"         bne-    .Lieaa%="
> @@ -292,7 +293,8 @@ g_atomic_int_add (volatile gint *atomic,
>  		  gint           val)
>  {
>    gint result, temp;
> -  __asm__ __volatile__ (".Lia%=:       lwarx   %0,0,%3\n"
> +  __asm__ __volatile__ ("nop\n"
> +			".Lia%=:       lwarx   %0,0,%3\n"
>  			"         add     %1,%0,%4\n"
>  			"         stwcx.  %1,0,%3\n"
>  			"         bne-    .Lia%="
>

I think you should rather report this upstream.

--
Anthony Ramine, the infamous MacPorts Trac slave.
nox at macports.org